Research And Optimization Of The Hybrid Distributed Storage System

Virtualization has become the foundation of cloud computing because it can maximize the hardware potential by running multiple operating systems and applications on a single physical machine.However,the traditional mechanical hard disk drives(HDD)has been unable to meet the high-speed data processing requirements of cloud computing.Thus,the non-volatile memory represented by the solid state drives(SSD)has been more and more popular.Compared with HDD,SSD provides a huge performance improvement.And with advances in technology,SSD capacity continues to increase and the prices are declining.But for data centers,storing all the data in SSD is still a high-cost solution.To solve this problem,hybrid storage systems of SSD and HDD represent a balanced solution between performance requirements and cost.This paper presents a hybrid storage solution,RHOBBS,which utilizes object storage to provide a virtual block storage device for virtual machines.In RHOBBS,according to the type of storage devices,all of the object storage device is divided into two logical storage pools,SSD pool and HHD pool.RHOBBS divides virtual machine disk images into fixed-size objects and each of which stores a portion of the virtual machine disk image.Based on object storage,RHOBBS provides a reliable,scalable virtual block storage device.RHOBBS uses the method of analyzing the contents of the virtual machine disk image to determine how the data objects are distributed among the different storage pools according to the contents of the data objects,thus effectively improving the IO performance of the virtual machine in the initial stage.Then,RHOBBS uses a cluster of shufflers to collect and analyze theworkload information from all the clients,and with the results of the analysis RHOBBS determines the global data object placement strategy.According to the placement strategy,data objects are migrated between the different storage pools.The cluster of shufflers will continue to work even if the minority of shufflers crash.Finally,in order to verify the effectiveness and correctness of the proposed solution,a simulation experiment is designed to show the performance of RHOBBS and the reliability of data migration function.The results show that RHOBBS improves the IO performance of virtual machines through data migration and virtual machine disk image analysis,and the cluster of shufflers can continue to work with a few nodes crashing.Our project has good contribution to cloud storage research.